distributed/protocol.c

   1 /* The functions implementing the master-slave protocol of the
   2  * distributed engine are grouped here. They are independent
   3  * of the gtp protocol. See the comments at the top of distributed.c
   4  * for a general introduction to the distributed engine. */
   5
   6 /* The receive queue is an array of pointers to binary buffers.
   7  * These pointers are invalidated in one of two ways when a buffer
   8  * is recycled: (1) the queue age is increased when the queue is
   9  * emptied at a new move, (2) the pointer itself is set to NULL
  10  * immmediately, and stays so until at least the next queue age
  11  * increment. */
  12
  13 #include <assert.h>
  14 #include <stdio.h>
  15 #include <pthread.h>
  16 #include <ctype.h>
  17 #include <unistd.h>
  18
  19 #define DEBUG
  20
  21 #include "random.h"
  22 #include "timeinfo.h"
  23 #include "playout.h"
  24 #include "network.h"
  25 #include "debug.h"
  26 #include "distributed/distributed.h"
  27 #include "distributed/protocol.h"
  28
  29 /* All gtp commands for current game separated by \n */
  30 static char gtp_cmds[CMDS_SIZE];
  31
  32 /* Latest gtp command sent to slaves. */
  33 static char *gtp_cmd = NULL;
  34
  35 /* Slaves send gtp_cmd when cmd_count changes. */
  36 static int cmd_count = 0;
  37
  38 /* Remember at most 10 gtp ids per move: kgs-rules, boardsize, clear_board,
  39  * time_settings, komi, handicap, genmoves, play pass, play pass, final_status_list */
  40 #define MAX_CMDS_PER_MOVE 10
  41
  42 /* History of gtp commands sent for current game, indexed by move. */
  43 static struct cmd_history {
  44         int gtp_id;
  45         char *next_cmd;
  46 } history[MAX_GAMELEN][MAX_CMDS_PER_MOVE];
  47
  48 /* Number of active slave machines working for this master. */
  49 int active_slaves = 0;
  50
  51 /* Number of replies to last gtp command already received. */
  52 int reply_count = 0;
  53
  54 /* All replies to latest gtp command are in gtp_replies[0..reply_count-1]. */
  55 char **gtp_replies;
  56
  57
  58 struct buf_state **receive_queue;
  59 int queue_length = 0;
  60 int queue_age = 0;
  61 static int queue_max_length;
  62
  63 /* Mutex protecting all variables above. receive_queue may be
  64  * read without the lock but is only written with lock held. */
  65 static pthread_mutex_t slave_lock = PTHREAD_MUTEX_INITIALIZER;
  66
  67 /* Condition signaled when a new gtp command is available. */
  68 static pthread_cond_t cmd_cond = PTHREAD_COND_INITIALIZER;
  69
  70 /* Condition signaled when reply_count increases. */
  71 static pthread_cond_t reply_cond = PTHREAD_COND_INITIALIZER;
  72
  73 /* Mutex protecting stderr. Must not be held at same time as slave_lock. */
  74 static pthread_mutex_t log_lock = PTHREAD_MUTEX_INITIALIZER;
  75
  76 /* Absolute time when this program was started.
  77  * For debugging only. */
  78 static double start_time;
  79
  80 /* Default slave state. */
  81 struct slave_state default_sstate;
  82
  83
  84 /* Get exclusive access to the threads and commands state. */
  85 void
  86 protocol_lock(void)
  87 {
  88         pthread_mutex_lock(&slave_lock);
  89 }
  90
  91 /* Release exclusive access to the threads and commands state. */
  92 void
  93 protocol_unlock(void)
  94 {
  95         pthread_mutex_unlock(&slave_lock);
  96 }
  97
  98 /* Write the time, client address, prefix, and string s to stderr atomically.
  99  * s should end with a \n */
 100 void
 101 logline(struct in_addr *client, char *prefix, char *s)
 102 {
 103         double now = time_now();
 104
 105         char addr[INET_ADDRSTRLEN];
 106         if (client) {
 107                 inet_ntop(AF_INET, client, addr, sizeof(addr));
 108         } else {
 109                 addr[0] = '\0';
 110         }
 111         pthread_mutex_lock(&log_lock);
 112         fprintf(stderr, "%s%15s %9.3f: %s", prefix, addr, now - start_time, s);
 113         pthread_mutex_unlock(&log_lock);
 114 }
 115
 116 /* Thread opening a connection on the given socket and copying input
 117  * from there to stderr. */
 118 static void *
 119 proxy_thread(void *arg)
 120 {
 121         int proxy_sock = (long)arg;
 122         assert(proxy_sock >= 0);
 123         for (;;) {
 124                 struct in_addr client;
 125                 int conn = open_server_connection(proxy_sock, &client);
 126                 FILE *f = fdopen(conn, "r");
 127                 char buf[BSIZE];
 128                 while (fgets(buf, BSIZE, f)) {
 129                         logline(&client, "< ", buf);
 130                 }
 131                 fclose(f);
 132         }
 133 }
 134
 135 /* Get a reply to one gtp command. Return the gtp command id,
 136  * or -1 if error. reply must have at least CMDS_SIZE bytes.
 137  * The ascii reply ends with an empty line; if the first line
 138  * contains "@size", a binary reply of size bytes follows the
 139  * empty line. @size is not standard gtp, it is only used
 140  * internally by Pachi for the genmoves command; it must be the
 141  * last parameter on the line.
 142  * *bin_size is the maximum size upon entry, actual size on return.
 143  * slave_lock is not held on either entry or exit of this function. */
 144 static int
 145 get_reply(FILE *f, struct in_addr client, char *reply, void *bin_reply, int *bin_size)
 146 {
 147         double start = time_now();
 148
 149         int reply_id = -1;
 150         *reply = '\0';
 151         if (!fgets(reply, CMDS_SIZE, f)) return -1;
 152
 153         /* Check for binary reply. */
 154         char *s = strchr(reply, '@');
 155         int size = 0;
 156         if (s) size = atoi(s+1);
 157         assert(size <= *bin_size);
 158         *bin_size = size;
 159
 160         if (DEBUGV(s, 2))
 161                 logline(&client, "<<", reply);
 162         if ((*reply == '=' || *reply == '?') && isdigit(reply[1]))
 163                 reply_id = atoi(reply+1);
 164
 165         /* Read the rest of the ascii reply */
 166         char *line = reply + strlen(reply);
 167         while (fgets(line, reply + CMDS_SIZE - line, f) && *line != '\n') {
 168                 if (DEBUGL(3))
 169                         logline(&client, "<<", line);
 170                 line += strlen(line);
 171         }
 172         if (*line != '\n') return -1;
 173
 174         /* Read the binary reply if any. */
 175         int len;
 176         while (size && (len = fread(bin_reply, 1, size, f)) > 0) {
 177                 bin_reply = (char *)bin_reply + len;
 178                 size -= len;
 179         }
 180         if (*bin_size && DEBUGVV(2)) {
 181                 char buf[1024];
 182                 snprintf(buf, sizeof(buf), "read reply %d+%d bytes in %.4fms\n",
 183                          (int)strlen(reply), *bin_size,
 184                          (time_now() - start)*1000);
 185                 logline(&client, "= ", buf);
 186         }
 187         return size ? -1 : reply_id;
 188 }
 189
 190 /* Send the gtp command to_send and get a reply from the slave machine.
 191  * Write the reply in buf which must have at least CMDS_SIZE bytes.
 192  * If *bin_size > 0, send bin_buf after the gtp command.
 193  * Return any binary reply in bin_buf and set its size in bin_size.
 194  * bin_buf is private to the slave and need not be copied.
 195  * Return the gtp command id, or -1 if error.
 196  * slave_lock is held on both entry and exit of this function. */
 197 static int
 198 send_command(char *to_send, void *bin_buf, int *bin_size,
 199              FILE *f, struct slave_state *sstate, char *buf)
 200 {
 201         assert(to_send && gtp_cmd && bin_buf && bin_size);
 202         strncpy(buf, to_send, CMDS_SIZE);
 203         bool resend = to_send != gtp_cmd;
 204
 205         pthread_mutex_unlock(&slave_lock);
 206
 207         if (DEBUGL(1) && resend)
 208                 logline(&sstate->client, "? ",
 209                         to_send == gtp_cmds ? "resend all\n" : "partial resend\n");
 210
 211         double start = time_now();
 212         fputs(buf, f);
 213
 214         if (*bin_size)
 215                 fwrite(bin_buf, 1, *bin_size, f);
 216         fflush(f);
 217
 218         if (DEBUGV(strchr(buf, '@'), 2)) {
 219                 double ms = (time_now() - start) * 1000.0;
 220                 if (!DEBUGL(3)) {
 221                         char *s = strchr(buf, '\n');
 222                         if (s) s[1] = '\0';
 223                 }
 224                 logline(&sstate->client, ">>", buf);
 225                 if (*bin_size) {
 226                         char b[1024];
 227                         snprintf(b, sizeof(b),
 228                                  "sent cmd %d+%d bytes in %.4fms\n",
 229                                  (int)strlen(buf), *bin_size, ms);
 230                         logline(&sstate->client, "= ", b);
 231                 }
 232         }
 233
 234         /* Reuse the buffers for the reply. */
 235         *bin_size = sstate->max_buf_size;
 236         int reply_id = get_reply(f, sstate->client, buf, bin_buf, bin_size);
 237
 238         pthread_mutex_lock(&slave_lock);
 239         return reply_id;
 240 }
 241
 242 /* Return the command sent after that with the given gtp id,
 243  * or gtp_cmds if the id wasn't used in this game. If a play command
 244  * has overwritten a genmoves command, return the play command.
 245  * slave_lock is held on both entry and exit of this function. */
 246 static char *
 247 next_command(int cmd_id)
 248 {
 249         if (cmd_id == -1) return gtp_cmds;
 250
 251         int last_id = atoi(gtp_cmd);
 252         int reply_move = move_number(cmd_id);
 253         if (reply_move > move_number(last_id)) return gtp_cmds;
 254
 255         int slot;
 256         for (slot = 0; slot < MAX_CMDS_PER_MOVE; slot++) {
 257                 if (cmd_id == history[reply_move][slot].gtp_id) break;
 258         }
 259         if (slot == MAX_CMDS_PER_MOVE) return gtp_cmds;
 260
 261         char *next = history[reply_move][slot].next_cmd;
 262         assert(next);
 263         return next;
 264 }
 265
 266 /* Allocate buffers for a slave thread. The state should have been
 267  * initialized already as a copy of the default slave state.
 268  * slave_lock is not held on either entry or exit of this function. */
 269 static void
 270 slave_state_alloc(struct slave_state *sstate)
 271 {
 272         for (int n = 0; n < BUFFERS_PER_SLAVE; n++) {
 273                 sstate->b[n].buf = malloc2(sstate->max_buf_size);
 274                 sstate->b[n].owner = sstate->thread_id;
 275         }
 276         if (sstate->alloc_hook) sstate->alloc_hook(sstate);
 277 }
 278
 279 /* Get a free binary buffer, first invalidating it in the receive
 280  * queue if necessary. In practice all buffers should be used
 281  * before they are invalidated, if BUFFERS_PER_SLAVE is large enough.
 282  * slave_lock is held on both entry and exit of this function. */
 283 static void *
 284 get_free_buf(struct slave_state *sstate)
 285 {
 286         int newest = (sstate->newest_buf + 1) & (BUFFERS_PER_SLAVE - 1);
 287         sstate->newest_buf = newest;
 288         void *buf = sstate->b[newest].buf;
 289
 290         if (DEBUGVV(7)) {
 291                 char b[1024];
 292                 snprintf(b, sizeof(b),
 293                          "get free %d index %d buf=%p age %d qlength %d\n", newest,
 294                          sstate->b[newest].queue_index, buf, queue_age, queue_length);
 295                 logline(&sstate->client, "? ", b);
 296         }
 297
 298         int index = sstate->b[newest].queue_index;
 299         if (index < 0) return buf;
 300
 301         /* Invalidate the buffer if the calling thread still owns its previous
 302          * entry in the receive queue. The entry may have been overwritten by
 303          * another thread, but only after a new move which invalidates the
 304          * entire receive queue. */
 305         if (receive_queue[index] && receive_queue[index]->owner == sstate->thread_id) {
 306                 receive_queue[index] = NULL;
 307         }
 308         sstate->b[newest].queue_index = -1;
 309         return buf;
 310 }
 311
 312 /* Insert a buffer in the receive queue. It should be the most
 313  * recent buffer allocated by the calling thread.
 314  * slave_lock is held on both entry and exit of this function. */
 315 static void
 316 insert_buf(struct slave_state *sstate, void *buf, int size)
 317 {
 318         assert(queue_length < queue_max_length);
 319
 320         int newest = sstate->newest_buf;
 321         assert(buf == sstate->b[newest].buf);
 322
 323         /* Update the buffer if necessary before making it
 324          * available to other threads. */
 325         if (sstate->insert_hook) sstate->insert_hook(buf, size);
 326
 327         if (DEBUGVV(7)) {
 328                 char b[1024];
 329                 snprintf(b, sizeof(b),
 330                          "insert newest %d age %d rq[%d]->%p owner %d\n",
 331                          newest, queue_age, queue_length, buf, sstate->thread_id);
 332                         logline(&sstate->client, "? ", b);
 333         }
 334         receive_queue[queue_length] = &sstate->b[newest];
 335         receive_queue[queue_length]->size = size;
 336         receive_queue[queue_length]->queue_index = queue_length;
 337         queue_length++;
 338 }
 339
 340 /* Clear the receive queue. The buffer pointers do not have to be cleared
 341  * here, this is done as each buffer is recycled.
 342  * slave_lock is held on both entry and exit of this function. */
 343 void
 344 clear_receive_queue(void)
 345 {
 346         if (DEBUGL(3)) {
 347                 char buf[1024];
 348                 snprintf(buf, sizeof(buf), "clear queue, old length %d age %d\n",
 349                          queue_length, queue_age);
 350                 logline(NULL, "? ", buf);
 351         }
 352         queue_length = 0;
 353         queue_age++;
 354 }
 355
 356 /* Process the reply received from a slave machine.
 357  * Copy the ascii part to reply_buf and insert the binary part
 358  * (if any) in the receive queue.
 359  * Return false if ok, true if the slave is out of sync.
 360  * slave_lock is held on both entry and exit of this function. */
 361 static bool
 362 process_reply(int reply_id, char *reply, char *reply_buf,
 363               void *bin_reply, int bin_size, int *last_reply_id,
 364               int *reply_slot, struct slave_state *sstate)
 365 {
 366         /* Resend everything if slave returned an error. */
 367         if (*reply != '=') {
 368                 *last_reply_id = -1;
 369                 return true;
 370         }
 371         /* Make sure we are still in sync. cmd_count may have
 372          * changed but the reply is valid as long as cmd_id didn't
 373          * change (this only occurs for consecutive genmoves). */
 374         int cmd_id = atoi(gtp_cmd);
 375         if (reply_id != cmd_id) {
 376                 *last_reply_id = reply_id;
 377                 return true;
 378         }
 379
 380         strncpy(reply_buf, reply, CMDS_SIZE);
 381         if (reply_id != *last_reply_id)
 382                 *reply_slot = reply_count++;
 383         gtp_replies[*reply_slot] = reply_buf;
 384
 385         if (bin_size) insert_buf(sstate, bin_reply, bin_size);
 386
 387         pthread_cond_signal(&reply_cond);
 388         *last_reply_id = reply_id;
 389         return false;
 390 }
 391
 392 /* Get the binary arg for the given command, and update the command
 393  * if necessary. For now, only genmoves has a binary argument, and
 394  * we return the best stats increments from all other slaves.
 395  * Set *bin_size to 0 if the command doesn't take binary arguments,
 396  * but still return a buffer, to be used for the reply.
 397  * Return NULL if the binary arg is obsolete by the time we have
 398  * finished computing it, because a new command is available.
 399  * This version only gets the buffer for the reply, to be completed
 400  * in future commits.
 401  * slave_lock is held on both entry and exit of this function. */
 402 void *
 403 get_binary_arg(struct slave_state *sstate, char *cmd, int cmd_size, int *bin_size)
 404 {
 405         int cmd_id = atoi(gtp_cmd);
 406         void *buf = get_free_buf(sstate);
 407
 408         *bin_size = 0;
 409         char *s = strchr(cmd, '@');
 410         if (!s || !sstate->args_hook) return buf;
 411
 412         int size = sstate->args_hook(buf, sstate, cmd_id);
 413
 414         /* Check that the command is still valid. */
 415         if (atoi(gtp_cmd) != cmd_id) return NULL;
 416
 417         /* Set the correct binary size for this slave.
 418          * cmd may have been overwritten with new parameters. */
 419         *bin_size = size;
 420         s = strchr(cmd, '@');
 421         assert(s);
 422         snprintf(s, cmd + cmd_size - s, "@%d\n", size);
 423         return buf;
 424 }
 425
 426 /* Main loop of a slave thread.
 427  * Send the current command to the slave machine and wait for a reply.
 428  * Resend command history if the slave machine is out of sync.
 429  * Returns when the connection with the slave machine is cut.
 430  * slave_lock is held on both entry and exit of this function. */
 431 static void
 432 slave_loop(FILE *f, char *reply_buf, struct slave_state *sstate, bool resend)
 433 {
 434         char *to_send;
 435         int last_cmd_count = 0;
 436         int last_reply_id = -1;
 437         int reply_slot = -1;
 438         for (;;) {
 439                 if (resend) {
 440                         /* Resend complete or partial history */
 441                         to_send = next_command(last_reply_id);
 442                 } else {
 443                         /* Wait for a new command. */
 444                         while (last_cmd_count == cmd_count)
 445                                 pthread_cond_wait(&cmd_cond, &slave_lock);
 446                         to_send = gtp_cmd;
 447                 }
 448
 449                 /* Command available, send it to slave machine.
 450                  * If slave was out of sync, send the history.
 451                  * But first get binary arguments if necessary. */
 452                 int bin_size = 0;
 453                 void *bin_buf = get_binary_arg(sstate, gtp_cmd,
 454                                                gtp_cmds + CMDS_SIZE - gtp_cmd,
 455                                                &bin_size);
 456                 /* Check that the command is still valid. */
 457                 resend = true;
 458                 if (!bin_buf) continue;
 459
 460                 /* Send the command and get the reply, which always ends with \n\n
 461                  * The slave machine sends "=id reply" or "?id reply"
 462                  * with id == cmd_id if it is in sync. */
 463                 last_cmd_count = cmd_count;
 464                 char buf[CMDS_SIZE];
 465                 int reply_id = send_command(to_send, bin_buf, &bin_size, f,
 466                                             sstate, buf);
 467                 if (reply_id == -1) return;
 468
 469                 resend = process_reply(reply_id, buf, reply_buf, bin_buf, bin_size,
 470                                        &last_reply_id, &reply_slot, sstate);
 471         }
 472 }
 473
 474 /* Minimimal check of slave identity. Close the file if error. */
 475 static bool
 476 is_pachi_slave(FILE *f, struct in_addr *client)
 477 {
 478         char buf[1024];
 479         fputs("name\n", f);
 480         if (!fgets(buf, sizeof(buf), f)
 481             || strncasecmp(buf, "= Pachi", 7)
 482             || !fgets(buf, sizeof(buf), f)
 483             || strcmp(buf, "\n")) {
 484                 logline(client, "? ", "bad slave\n");
 485                 fclose(f);
 486                 sleep(1); // avoid busy loop if error
 487                 return false;
 488         }
 489         return true;
 490 }
 491
 492 /* Thread sending gtp commands to one slave machine, and
 493  * reading replies. If a slave machine dies, this thread waits
 494  * for a connection from another slave.
 495  * The large buffers are allocated only once we get a first
 496  * connection, to avoid wasting memory if max_slaves is too large.
 497  * We do not invalidate the received buffers if a slave disconnects;
 498  * they are still useful for other slaves. */
 499 static void *
 500 slave_thread(void *arg)
 501 {
 502         struct slave_state sstate = default_sstate;
 503         sstate.thread_id = (long)arg;
 504
 505         assert(sstate.slave_sock >= 0);
 506         char reply_buf[CMDS_SIZE];
 507         bool resend = false;
 508
 509         for (;;) {
 510                 /* Wait for a connection from any slave. */
 511                 struct in_addr client;
 512                 int conn = open_server_connection(sstate.slave_sock, &client);
 513
 514                 FILE *f = fdopen(conn, "r+");
 515                 if (DEBUGL(2)) {
 516                         snprintf(reply_buf, sizeof(reply_buf),
 517                                  "new slave, id %d\n", sstate.thread_id);
 518                         logline(&client, "= ", reply_buf);
 519                 }
 520                 if (!is_pachi_slave(f, &client)) continue;
 521
 522                 if (!resend) slave_state_alloc(&sstate);
 523                 sstate.client = client;
 524
 525                 pthread_mutex_lock(&slave_lock);
 526                 active_slaves++;
 527                 slave_loop(f, reply_buf, &sstate, resend);
 528
 529                 assert(active_slaves > 0);
 530                 active_slaves--;
 531                 // Unblock main thread if it was waiting for this slave.
 532                 pthread_cond_signal(&reply_cond);
 533                 pthread_mutex_unlock(&slave_lock);
 534
 535                 resend = true;
 536                 if (DEBUGL(2))
 537                         logline(&client, "= ", "lost slave\n");
 538                 fclose(f);
 539         }
 540 }
 541
 542 /* Create a new gtp command for all slaves. The slave lock is held
 543  * upon entry and upon return, so the command will actually be
 544  * sent when the lock is released. The last command is overwritten
 545  * if gtp_cmd points to a non-empty string. cmd is a single word;
 546  * args has all arguments and is empty or has a trailing \n */
 547 void
 548 update_cmd(struct board *b, char *cmd, char *args, bool new_id)
 549 {
 550         assert(gtp_cmd);
 551         /* To make sure the slaves are in sync, we ignore the original id
 552          * and use the board number plus some random bits as gtp id. */
 553         static int gtp_id = -1;
 554         int moves = is_reset(cmd) ? 0 : b->moves;
 555         if (new_id) {
 556                 int prev_id = gtp_id;
 557                 do {
 558                         /* fast_random() is 16-bit only so the multiplication can't overflow. */
 559                         gtp_id = force_reply(moves + fast_random(65535) * DIST_GAMELEN);
 560                 } while (gtp_id == prev_id);
 561                 reply_count = 0;
 562         }
 563         snprintf(gtp_cmd, gtp_cmds + CMDS_SIZE - gtp_cmd, "%d %s %s",
 564                  gtp_id, cmd, *args ? args : "\n");
 565         cmd_count++;
 566
 567         /* Remember history for out-of-sync slaves. */
 568         static int slot = 0;
 569         static struct cmd_history *last = NULL;
 570         if (new_id) {
 571                 if (last) last->next_cmd = gtp_cmd;
 572                 slot = (slot + 1) % MAX_CMDS_PER_MOVE;
 573                 last = &history[moves][slot];
 574                 last->gtp_id = gtp_id;
 575                 last->next_cmd = NULL;
 576         }
 577         // Notify the slave threads about the new command.
 578         pthread_cond_broadcast(&cmd_cond);
 579 }
 580
 581 /* Update the command history, then create a new gtp command
 582  * for all slaves. The slave lock is held upon entry and
 583  * upon return, so the command will actually be sent when the
 584  * lock is released. cmd is a single word; args has all
 585  * arguments and is empty or has a trailing \n */
 586 void
 587 new_cmd(struct board *b, char *cmd, char *args)
 588 {
 589         // Clear the history when a new game starts:
 590         if (!gtp_cmd || is_gamestart(cmd)) {
 591                 gtp_cmd = gtp_cmds;
 592                 memset(history, 0, sizeof(history));
 593         } else {
 594                 /* Preserve command history for new slaves.
 595                  * To indicate that the slave should only reply to
 596                  * the last command we force the id of previous
 597                  * commands to be just the move number. */
 598                 int id = prevent_reply(atoi(gtp_cmd));
 599                 int len = strspn(gtp_cmd, "0123456789");
 600                 char buf[32];
 601                 snprintf(buf, sizeof(buf), "%0*d", len, id);
 602                 memcpy(gtp_cmd, buf, len);
 603
 604                 gtp_cmd += strlen(gtp_cmd);
 605         }
 606
 607         // Let the slave threads send the new gtp command:
 608         update_cmd(b, cmd, args, true);
 609 }
 610
 611 /* Wait for at least one new reply. Return when at least
 612  * min_replies slaves have already replied, or when the
 613  * given absolute time is passed.
 614  * The replies are returned in gtp_replies[0..reply_count-1]
 615  * slave_lock is held on entry and on return. */
 616 void
 617 get_replies(double time_limit, int min_replies)
 618 {
 619         for (;;) {
 620                 if (reply_count > 0) {
 621                         struct timespec ts;
 622                         double sec;
 623                         ts.tv_nsec = (int)(modf(time_limit, &sec)*1000000000.0);
 624                         ts.tv_sec = (int)sec;
 625                         pthread_cond_timedwait(&reply_cond, &slave_lock, &ts);
 626                 } else {
 627                         pthread_cond_wait(&reply_cond, &slave_lock);
 628                 }
 629                 if (reply_count == 0) continue;
 630                 if (reply_count >= min_replies || reply_count >= active_slaves) return;
 631                 if (time_now() >= time_limit) break;
 632         }
 633         if (DEBUGL(1)) {
 634                 char buf[1024];
 635                 snprintf(buf, sizeof(buf),
 636                          "get_replies timeout %.3f >= %.3f, replies %d < min %d, active %d\n",
 637                          time_now() - start_time, time_limit - start_time,
 638                          reply_count, min_replies, active_slaves);
 639                 logline(NULL, "? ", buf);
 640         }
 641         assert(reply_count > 0);
 642 }
 643
 644 /* In a 5mn move with at least 5ms per genmoves we get at most
 645  * 300*200=60000 genmoves per slave. */
 646 #define MAX_GENMOVES_PER_SLAVE 60000
 647
 648 /* Allocate the receive queue, and create the slave and proxy threads.
 649  * max_buf_size and the merge-related fields of default_sstate must
 650  * already be initialized. */
 651 void
 652 protocol_init(char *slave_port, char *proxy_port, int max_slaves)
 653 {
 654         start_time = time_now();
 655
 656         queue_max_length = max_slaves * MAX_GENMOVES_PER_SLAVE;
 657         receive_queue = calloc2(queue_max_length, sizeof(*receive_queue));
 658
 659         default_sstate.slave_sock = port_listen(slave_port, max_slaves);
 660         default_sstate.last_processed = -1;
 661
 662         for (int n = 0; n < BUFFERS_PER_SLAVE; n++) {
 663                 default_sstate.b[n].queue_index = -1;
 664         }
 665
 666         pthread_t thread;
 667         for (int id = 0; id < max_slaves; id++) {
 668                 pthread_create(&thread, NULL, slave_thread, (void *)(long)id);
 669         }
 670
 671         if (proxy_port) {
 672                 int proxy_sock = port_listen(proxy_port, max_slaves);
 673                 for (int id = 0; id < max_slaves; id++) {
 674                         pthread_create(&thread, NULL, proxy_thread, (void *)(long)proxy_sock);
 675                 }
 676         }
 677 }